Top Rated Pune Call Girls Dighi ⟟ 6297143586 ⟟ Call Me For Genuine Sex Servi...
Smart fundmgrsstupidmoney bernhardt
1. Smart fund managers? Stupid money?
Dan Bernhardt Department of Economics, University of Illinois
Ryan J. Davies Finance Division, Babson College
Abstract. We develop a model of mutual fund manager investment decisions near the end
of quarters. We show that when investors reward better performing funds with higher cash
flows, near quarter-ends a mutual fund manager has an incentive to distort new invest-
ment toward stocks in which his fund holds a large existing position. The short-term price
impact of these trades increase the fund’s reported returns. Higher returns are rewarded
by greater subsequent fund inflows which, in turn, allow for more investment distortion
the next quarter. Because the price impact of trades is short term, each subsequent quar-
ter begins with a larger return deficit. Eventually, the deficit cannot be overcome. Thus,
our model leads to the empirically observed short-run persistence and long-run rever-
sal in fund performance. In doing so, our model provides a consistent explanation of
many other seemingly contradictory empirical features of mutual fund performance. JEL
classification: D82, G2, G14
Malins gestionnaires de fonds ? Investisseurs stupides ? Les auteurs d´ veloppent un mod` le
e e
de d´ cisions d’investissement des gestionnaires de fonds mutuels pr` s de la fin d’un
e e
trimestre. On montre que quand les investisseurs r´ compensent les fonds mutuels per-
e
formants en y injectant des ressources additionnelles, vers la fin d’un trimestre, le gestion-
`
naire de fonds a une incitation a infl´ chir le nouvel investissement vers des actions dans
e
lesquelles le fond a une forte position existante. L’impact de ces transactions sur les prix
`
de ces actions a court terme augmente les rendements trimestriels rapport´ s pour le fond.
e
De meilleurs rendements sont r´ compens´ s par un influx d’investissement subs´ quent, ce
e e e
`
qui ouvre la porte a encore plus de d´ flection d’investissement au trimestre suivant. Parce
e
Richard Martin’s master’s essay under the direction of the first author was a building point for
this research. We greatly appreciate Harvey Westbrook Jr’s significant contributions to early
drafts. We are also grateful for comments received from two anonymous referees, Chris Brooks,
Murillo Campello, Duane Seppi, and seminar participants at Queen’s University, St Francis
Xavier University, the Northern Finance Association meetings, the Financial Management
Association European meetings, and the Multinational Finance Society meetings.
Email: rdavies@babson.edu; danber@uiuc.edu
Canadian Journal of Economics / Revue canadienne d’Economique, Vol. 42, No. 2
May / mai 2009. Printed in Canada / Imprim´ au Canada
e
0008-4085 / 09 / 719–748 / Canadian Economics Association
C
2. 720 D. Bernhardt and R.J. Davies
`
que l’impact de ces transactions sur le prix des actions est a court terme, chaque trimestre
subs´ quent commence avec un d´ ficit de rendements plus important. Eventuellement, il
e e
n’est plus possible de combler ce d´ ficit. Ce mod` le permet d’expliquer les observations
e e
`
empiriques de persistance de mesures de performance du fond a court terme et de ren-
`
versement de la tendance a long terme. Le mod` le fournit aussi une explication consistante
e
de plusieurs autres caract´ ristiques apparemment contradictoires not´ es empiriquement
e e
dans la performance des fonds mutuels.
1. Introduction
In this paper, we develop a model of mutual fund manager investment de-
cisions near the end of quarters. Our basic intuition can be summarized as
follows. We begin with the well-established observation that investors reward
better-performing mutual funds with greater cash inflows. Small improvements in
relative performance can generate large increases in future fund cash inflows. 1
Thus, mutual fund managers, whose compensation rises with assets under man-
agement, have strong incentives to increase the reported quarterly returns of
their funds. We show that one method for doing so is for the fund to distort its
end-of-quarter purchases toward stocks in which the fund already holds large po-
sitions. The short-term price impact of these trades increases the reported value
of the fund’s existing positions, thereby raising the fund’s end-of-quarter reported
returns.
A fund’s ability to engage in this distortive end-of-quarter trading is lim-
ited by the amount of new cash it has available to invest. Funds with bet-
ter past performance receive larger cash inflows and this facilitates more ag-
gressive end-of-quarter trading. Of course, this distortion comes at a cost –
there is no free lunch. Specifically, the price impacts of these trades eventu-
ally decay, decreasing the next quarter’s return, all else equal. However, over
time, to overcome the hangover from past distortive trading, funds must en-
gage in ever more distortive trades. Eventually the accumulated hangover proves
too much, leading to the long-run reversals in fund performance found in the
data.
Thus, our model provides a framework for explaining the puzzling empirical
evidence that mutual fund returns exhibit short-run persistence (‘hot and cold
hands’) and long-run reversals (past top performing funds become future un-
derperformers, even after incorporating their initially superior performance). 2 In
addition, the cost of the investment distortions that we model can help to explain
1 For example, see Chevalier and Ellison (1997), Ippolito (1992), and Sirri and Tufano (1998).
2 Extensive evidence of short-term persistence and long-run reversals in the performance of
mutual funds is provided by Hendricks, Patel, and Zeckhauser (1993), Malkiel (1995), Brown
and Goetzmann (1995), Gruber (1996), Carhart (1997), Zheng (1999), Wermers (2003), and
Bollen and Busse (2005).
3. Smart fund managers? Stupid money? 721
why mutual funds tend to underperform non-managed indexes, despite evidence
that mutual fund managers have some stock picking ability. 3
A key component of our model is that the price impact of trades decays over
time. Early in a quarter, fund managers may trade so as to minimize investment
distortions/price impacts. Later in the quarter, however, enough of the price
impact persists through the end of the quarter to make it attractive to distort
investments toward assets in place. An extreme manifestation of this is the practice
of ‘high closing’ – artificially boosting the price of a stock by purchasing shares
just before the market closes on the last day of the quarter. 4
Trades in less liquid stocks generally have greater price impacts. Thus, our
model predicts that managers of funds that specialize in less liquid stocks have
greater incentives to distort end-of-quarter investment toward existing holdings.
Over time, this behaviour will lead these funds to hold undiversified portfo-
lios. This is consistent with the observed high-return volatility of specialized
funds. Our model’s link between short-run fund performance and the concentra-
tion of fund portfolios can help to reconcile the findings of Kacperczyk, Sialm,
and Zheng (2005) on the performance impact of industry concentration in fund
holdings.
Our model suggests that funds may adapt their overall holdings to exploit
the incentives modelled here. For instance, smaller funds may focus more on
smaller, illiquid stocks. In this way, a small fund can be a ‘big fish in a small
pond.’ While the typical trade size of a small fund is too small to have much
influence on the price of a large stock, the same trade size can have a large
impact on prices of less liquid stocks. Similarly, our model predicts that, since
younger funds have a more sensitive performance-flow relationship, managers
of younger funds have greater incentives to distort investment toward assets in
place. 5 Thus, our model predictions are consistent with the empirical findings
of Zheng (1999) and Wermers (2003) that smaller funds exhibit greater short-
run persistence in performance and even more dramatic long-run reversals in
performance.
Our model also suggests that mutual funds will tend to be momentum traders –
when a mutual fund’s past superior performance is generated by large positions
in well-performing stocks, our model predicts that the mutual fund will continue
to increases its positions in those stocks, endogenously generating momentum
3 Wermers, Chen, and Jegadeesh (2000) and Pinnuck (2003) observe that stocks purchased by
mutual funds tend to outperform stocks that they sell; yet, on average, actively managed funds
underperform index funds.
4 John Gilfoyle reports, ‘Nearly everyone seems to agree that high closing is common.’ (Macleans,
10 July 2000, 39). For example, the Ontario Securities Commission in its case against RT Capital
Management, Inc. cites the end of year purchase of 1900 shares of Dia Met at a $.50 premium.
A Globe and Mail (7 July 2000) investigation found that in mutual funds on the final trading day
of the year, ‘last-minute leaps beyond the normal market trends strongly suggested a round of
“portfolio pumping.”’
5 Chevalier and Ellison (1997) find that younger funds must perform better to attract investment.
4. 722 D. Bernhardt and R.J. Davies
through the resulting price impact. Our model also predicts that there will be a
long-run reversal in performance, that is, that those subsequent purchases will
earn low long-run returns. San (2007) documents each of these empirical regu-
larities for trading by institutions.
Our model takes investors’ decisions to reward higher short-term returns with
larger cash inflows as given. There is ample empirical evidence documenting this
investor behaviour. Our analysis suggests that the best response of investors is
not given by this stimulus-response setup – hence the reference to ‘stupid money’
in the title; instead, investors’ optimal investment decisions would need to reflect
the strategic incentives of the fund managers. To endogenize the responses of
individual investors, we would need to extend our model to include heterogeneity
in fund manager ability and integrate learning by investors from mutual fund
returns. Bernhardt and Nosal (2008) do this in a strategically simpler setting
where a hedge fund manager can augment fund payoffs with zero NPV gambles
of his own design. In our strategically richer setting, it is infeasible to endogenize
both investor and fund manager behaviour, and the reader should be alert to the
possibility that conclusions about ‘stupid money’ might be changed in such a
setting.
The paper proceeds as follows. The next section provides an overview of re-
lated literature. Section 3 sets up the economic environment. Section 4 analyzes
how a fund’s existing stock holdings influence its manager’s investment decisions
and derive the consequences of these decisions for persistence in fund returns.
Section 5 provides numerical examples. Section 6 concludes. Proofs are in an
appendix.
2. Related literature
Gallagher, Gardner, and Swan (2007) use daily trading data of investment man-
agers to provide direct evidence of ‘painting the tape’ and show that gaming
behaviour is more likely to occur in smaller stocks, growth stocks, and stocks in
which the fund has a larger position.
Carhart et al. (2002) provide further extensive empirical support. They find
that funds earn tremendous short-run returns near the end of an evaluation
period, when trading behaviour has the greatest impact on performance: 80%
of funds beat the S&P 500 Index on the last trading day of the year (62% for
other quarter-end dates), but only 37% (40% other quarters) do so on the first
trading day of a new year. This effect is even greater for small-cap funds, which
trade less liquid stocks: 91% of small-cap funds beat the index at year’s end,
and 70% for other quarter-end dates versus 34% for first-quarter trading day.
Carhart et al. (2002) reject benchmark-beating hypotheses in favour of strategic
behaviour similar to that motivated here. Consistent with our theory, they find
funds that performed better in the past year earned 42 basis points higher returns
on the last trading day and 29 basis points lower on the first trading day than
5. Smart fund managers? Stupid money? 723
funds with a worse historical performance. In sum, the data strongly support the
hypothesis that fund managers respond to the short-run trading incentives that
we model, and that they are collectively substantial enough to alter aggregate
market outcomes.
Indeed, Bernhardt and Davies (2005) find that strategic trading by fund man-
agers appears to impact returns on aggregate market indexes, so that measuring
the impact relative to the index, as Carhart et al. (2002) do, understates the total
effect. Specifically, daily returns of the equally weighted index on the last trading
day of a quarter greatly exceed the daily returns on the first trading day of the
succeeding quarter, and this return difference rises with the share of total equity
held by mutual funds.
Zheng (1999) finds that funds with relatively high returns in one quarter have
significantly greater returns than the average mutual fund in the next quarter,
and relatively worse performers in one quarter generate lower returns than the
average mutual fund in the next period. Zheng (1999) shows that the performance
persistence is short lived: more than one-quarter into the future, funds that did
better in the past under perform relative to the average fund, while historically
poor-performing funds, do better. This reversal in performance is so strong that
cumulative returns of historically better performers fall below those of worse
performers within 30 months.
At the stock-holdings level, Wermers (2003) finds that funds ranked in the top
quintile in the prior year (‘past winners’) beat the S&P 500 by 2% in the next year
(after accounting for trading costs and expenses). He also finds that past winners
beat past losers (funds in the prior year’s bottom quintile) by 5% in the next
year. Wermers (2003) provides empirical support for our theoretical connection
between fund flows and trades, showing that flow-related buying drives stock
prices up, and that this flow-related buying plays a central role in driving the
persistence in fund performance.
In an insightful paper, Berk and Green (2004) develop a model in which
rational investors learn from a fund’s performance about its manager’s abil-
ity and allocate funds accordingly. The Berk-Green model can explain both
why money flows into mutual funds with recent better performance and away
from those with worse past performance. However, the Berk-Green model also
implies that past returns should not predict future performance, which is in-
consistent with short-term performance persistence and long-run performance
reversals found in the data. Primary contributions of our model are to rec-
oncile these empirical observations and to provide explanations for several
features of the mutual fund market not explained by the Berk-Green model,
such as why stocks purchased by mutual funds outperform those that they
sell, yet, on average, mutual funds underperform the market; and the end-
of-quarter return patterns and strategic behaviour documented by Gallagher,
Gardner, and Swan (2007), Carhart et al. (2002), and Bernhardt and Davies
(2005).
6. 724 D. Bernhardt and R.J. Davies
3. The model
3.1. Model details
Consider a risk-neutral mutual fund manager who can invest in three assets: stock
A, stock B, and cash. The fund manager enters quarter t with an existing position
of SAt shares in stock A and SBt shares in stock B. The associated share prices are
P At and P Bt . Without loss of generality we assume that PAt SAt ≥ PBt SBt . Cash
earns a risk-free return of i, which we normalize to zero.
Firms retain their earnings. Hence, a firm’s stock price is equal to its expected
discounted lifetime cash flows. We impose no structure on the timing of cash
flows. At the beginning of quarter t, each firm makes an earnings announcement
and provides guidance about future cash flows, which causes market participants
to update about firm values. We summarize the earnings announcement and
guidance by its implications for the percentage change in firm value, δ j,t , so that
δ j,t P j,t−1 is the change in expected discounted cash flows. 6 Following a signal of
δ j,t , investors update about cash flows, leading to a quarter t stock price for firm
j of
Pj,t = Pj,t−1 + δ j,t Pj,t−1 . (1)
We assume that in quarter t the fund manager privately learns the next quarter’s
signals, δ A,t+1 and δ B,t+1 , and can trade based on this private information in
quarter t before the signals are publicly revealed with certainty at the beginning
of quarter t + 1. The joint density distribution of signals across firms is given by
g(δ A , δ B ). Thus, in our model, mutual funds are informed traders, whose fully
rational trading decisions can be based on private information, in addition to
the liquidity needs arising from net cash inflows and the strategic consideration
of the price impacts of trades on returns. One role of private information in our
model is to show that incentives to pump existing holdings can be so strong
that the fund manager actually trades in the opposite direction of his private
information.
At the end of quarter t, the fund receives net cash inflows of f (r t ), where r t was
the fund’s portfolio return over quarter t. The only structure that we impose is that
f (r t ) is increasing in r t – greater past fund performance raises net cash inflows
from investors. Presumably, the performance-flow relation reflects investor beliefs
that fund managers differ in their abilities to identify good investments (Berk and
Green 2004). Here, we do not distinguish fund managers by ability, because such
differences can also lead to persistence in performance.
Consistent with the findings of Chan and Lakonishok (1995) and Keim and
Madhavan (1997), we assume that a fund manager’s stock purchases have short-
term price impacts: the more shares that a fund manager purchases, the greater
6 If δ j,t+1 is not proportional to firm value, then pricing is sensitive to a firm’s choice of shares
outstanding.
7. Smart fund managers? Stupid money? 725
is the short-term price impact. Specifically, a purchase of Ijt > 0 shares in stock j
costs Pjt + Pjt (Pjt , Ijt ) > Pjt per share. We summarize the properties of the price
impact of Ijt , Pjt (Pjt , Ijt ), later in assumptions A1–A4. We are agnostic as to the
fundamentals driving the price impact of Ijt – it could be the fact that institutional
trades contain information, or that market makers must be compensated for hav-
ing to readjust their portfolios, and it takes time to do so. What is key is only that
such institutional trading activity has a significant and persistent impact on price.
It is important to note that our pricing relation does not preclude the possibility
that market makers adjust their pricing to reflect the end-of-quarter incentives
of funds; 7 or that larger orders impose greater inventory costs on market makers. 8
In each quarter t, the timing of events is as follows:
1. Firms announce period earnings and market participants revise expectations
about the discounted present value of a firm’s cash flows so that stock prices
equal P A,t and P B,t .
2. The fund manager learns δ A,t+1 and δ B,t+1 and invests available cash to max-
imize the fund’s expected period return. Available cash consists of net cash
inflows f (r t−1 ) plus the present value of last period’s cash position Mt−1 . The
fund manager can sell shares, but cannot sell shares short or borrow to finance
stock investments.
3. The fund manager’s private information about next period’s signal, δ j,t+1 , is
revealed with (independent) probability, γ , and is not revealed with residual
probability (1 − γ ). We assume that γ ∈ (0, 1), that is, information is only
sometimes revealed to the public. A smaller value of γ reflects less leakage of
information between the purchase and the end of the period. Thus, γ captures
either (i) the length of time between the purchase and the end of the quarter
or (ii) the probability that private signals will be revealed (because the stock is
smaller and hence is followed by fewer analysts) or (iii) a combination of both
effects.
4. End-of-quarter stock prices, P∗ , P∗ , are realized. If the fund manager’s
At Bt
private information was revealed,
Pj∗t = Pj t + δ j,t+1 Pj t .
If the private information was not revealed,
Pj∗t = Pj t + Pj t (Pj t , I j t ).
7 Bhattacharyya and Nanda (2007) augment a standard single asset static Kyle (1985) model by
having an informed agent’s objective be the weighted average of (i) a short-run return based on
the price generated by his trade on his holdings of the asset and (ii) the final return on the stock,
when market maker’s have a signal about the agent’s pre-trade holdings of the asset. They derive
a linear pricing rule that is less sensitive to order flow than the standard model.
8 Hendershott and Seasholes (2007) document that, on average, prices rise when larger orders
substantially reduce market-maker inventories, before falling subsequently in the next week
when market makers re-establish their inventories.
8. 726 D. Bernhardt and R.J. Davies
FIGURE 1 Timeline of events in quarter t and the evolution of stock prices
5. End-of-quarter fund returns (r t ) are realized. The fund receives net cash in-
flows of f (r t ).
Figure 1 provides a sketch of the timing of events.
We next set out the sole structure imposed on the short-term price impact of
share trades:
A1. If Ijt = 0, there is no price impact: Pjt (Pjt , 0) = 0.
A2. The price impact of larger orders is greater: ∂ Pjt (Pjt , Ijt )/∂ Ijt > 0.
A3. The price schedules are symmetric: P At (·) = P Bt (·).
A4. The price schedule is concave, but not too concave:
∂ 2 Pj t (Pj t , I j t ) ∂ Pj t (Pj t , I j t ) ∂ 2 Pj t (Pj t , I j t )
≤ 0; 2 + Ijt > 0.
∂ I j2t ∂ Ijt ∂ I j2t
Assumption A1 is largely a normalization. Assumption A2 captures the empir-
ical regularity that larger orders have greater price impacts and is an equilib-
rium property of economic models in which individual orders are information-
ally large (Kyle 1985, etc.) or are large from the perspective of market-maker
inventories. Assumption A3 allows us to abstract away from how different price
schedules affect a fund manager’s investment decisions, and is consistent with the
assets’ having the same stochastic properties and market makers’ not knowing
the holdings of individual fund managers. Later, we allow price schedules to differ
across stocks. Assumption A4 is a technical assumption whose role is to ensure
that the fund manager’s optimal trading strategy is characterized by first-order
conditions.
Some of our results will be most transparent when we assume
A5. The price impact of an order is proportional to the value of the order:
P(Pj t , I j t ) = k(Pj t I j t )Pj t , k > 0. (2)
Assumption A5 is implied if pricing is neutral with respect to a firm’s choice of
shares outstanding.
9. Smart fund managers? Stupid money? 727
Fund manager’s problem. At the beginning of quarter t, the fund manager invests
to maximize the expected quarter portfolio return:
Mt + Sj,t+1 E[Pj∗t ]
j
max E[rt ] = ∗
− 1, (3)
Mt ,I At ,IBt Mt−1 + f (rt−1 ) + Pj,t−1 Sj t
j
subject to
Mt ≥ 0,
I j t ≥ −Sj t , j = A, B
[Pj t + Pj t (Pj t , I j t )]I j t + Mt ≤ f (rt−1 ) + Mt−1 ,
j
where Sj,t+1 = Sjt + Ijt and E[P∗ ] = Pjt + (1 − γ ) Pjt (Pjt , Ijt ) + γ δ j,t+1 Pjt .
jt
We will assume that the short-selling constraint, Ijt ≥ − Sjt , j = A, B does not
bind: the fund manager does not receive such a bad signal about a stock that he
wants to sell more shares of the stock than he has in his portfolio.
3.2. Discussion of the model
3.2.1. Modelling approach
The logical construction of our model is deceivingly simple, masking the fact that
it contains many ingredients that researchers have not yet been able to analyze
even separately within a structural market microstructure setting. Central to our
model is an informed fund manager who is budget constrained, chooses how
many shares of multiple assets to purchase or sell, makes investment decisions
over time, and must incorporate the fund’s existing asset holdings into trading
decisions. The existing holdings enter non-neutrally because the price impacts
of trades affect the portfolio return that investors observe and condition future
mutual fund investments on. Further complicating this dynamic optimization
problem, a fund manager’s trades must have price impacts, trading scales cannot
be trivial (i.e., zero versus single round lots), and trading rules will be highly
non-linear.
There is a large market microstructure literature dating back to Kyle (1985)
that analyzes a single asset that generates endogenously the price impacts of orders
of the form that we model and that are found in the data. 9 In Kyle and its multi-
agent extensions, risk-neutral speculators receive normally distributed signals
about one asset’s terminal value, submit orders over time that are mixed in with
normally distributed exogenous ‘noise trade,’ and a market maker sets price equal
to the asset’s expected value given the net order flow. The normal assumptions
imply linear conditional forecasts and hence linear pricing, making the problem
9 Noisy rational expectations frameworks are inappropriate here, as they assume that individuals
are small, so their individual trades do not generate the price impact found in the data.
10. 728 D. Bernhardt and R.J. Davies
solvable. Caball´ and Krishnan (1994) extend this analysis to multiple stocks in a
e
static setting and prove that a linear equilibrium exists. Bernhardt and Taub (2008)
provide rich analytical characterizations of the Caball´ and Krishnan model, and
e
they also solve for equilibrium outcomes when speculators can condition their
trades on prices. To our knowledge, there are no other theoretical papers with
multiple assets and price impacts.
Technically, the biggest challenge to handle in a model with full primitives is
the budget constraint, which destroys linearity of forecasts (and hence prices) in
models with normally distributed signals. Not only does our fund manager have
a budget constraint, but the budget is endogenous, there are multiple risky assets,
and assets-in-place critically affect optimal trades. While the budget constraint is
central to the economics of our results, it is not central to the form of pricing – in
any setting where an agent’s trade is ‘large,’ the equilibrium pricing schedule will
have price impacts (larger orders receive higher prices), simply because an agent
with a better signal about an asset buys more and must face an opportunity cost
of buying more – in the form of a higher price.
One can also derive endogenously the qualitative consequences of introduc-
ing an investor who wants to devote resources toward assets in place within an
equilibrium model. Bhattacharyya and Nanda (2007; hereafter B-N) do this by
adapting Kyle (1985) to consider a single investor who cares about both the in-
terim payoff on an existing inventory position on a single risky asset and the final
payoff. The investor has a two-period horizon. At date 1, the investor receives
a private signal and trades once in a simultaneous batch auction. At date 2, the
asset is liquidated at its true value. Thus, the investor’s date 1 trading decision
is based on the NAV of his position at dates 1 and 2; in particular, the investor
incorporates the price impact of trading on the value of his existing position at
date 1. As a result, the optimal trade size is linearly related to the size of the
existing position.
In the most general B-N setting, the market maker has a noisy signal of the
investor’s (normally distributed) existing position. This preserves linear forecasts
and hence linear pricing. The investor’s incentives to pump up the value of his
holdings reduce the sensitivity of the equilibrium price schedule to order flow,
but the qualitative properties of pricing are otherwise preserved. B-N argue that
the investor’s incentives to enhance short-term values can explain why closed-
end funds trade at a discount to their NAVs. Their model, however, cannot shed
light on any of the issues we study, as it cannot deal with multiple risky assets,
the budget constraints of mutual funds, or fund managers with multiple trading
opportunities.
In sum, while some of our assumptions are not grounded in primitive founda-
tions, our framework requires only a mild reduced-form structure on pricing. This
pricing structure holds empirically and allows us to get at the key links between
the market structure of asset pricing, strategic fund manager behaviour, and port-
folio returns, thereby reconciling a host of fundamental empirical regularities in
fund performance.
11. Smart fund managers? Stupid money? 729
3.2.2. Empirical evidence in support of price impact function
Keim and Madhavan (1996) find that the price impact of a block trade,
measured as the market-adjusted difference between the closing price on the
day prior to the trade and the closing price on the day after the trade, is
−1.50% for seller-initiated blocks and 1.60% for buyer-initiated blocks. Chan and
Lakonishok (1995) measure the price impact of a package of institutional trades
and find that buy packages are associated with a principal-weighted average price
change of almost 1% from the open on the package’s first day to the close on
the last day. The analogous price change for sell packages is −0.3%. This price
impact is persistent – for example, five days after the completion of a package
trade, there is a reversal in returns of only −0.07% for buy packages and 0.10%
for sell packages.
Frequently, institutional traders use crossing networks, such as POSIT, or use
so-called dark liquidity pools to reduce the price impact of their trades. It would
be interesting to investigate empirically whether mutual funds reduce their usage
of these alternative trading venues near the end-of-quarters.
3.2.3. Time horizon of fund manager
While fund manager compensation is proportional to assets under management, a
fund manager’s expected compensation may depend more on gaining a promotion
based on past fund performance. 10 Fund manager turnover is high: Baks (2007)
finds that on average about 45% of managers leave or start at a fund in any given
year and that the average time spent at one fund is about three years (less for
small company growth funds). To capture these incentives, we suppose that in each
period a fund manager invests to maximize end-of-period return. We interpret the
period’s end as the date at which investors receive fund performance information,
which leads them to reallocate investments. Focusing on one-period horizons for
fund managers eases the analysis, and the incentive effects highlighted remain
present with longer horizons.
The one-period horizon in our model is best interpreted as a quarter, since this
is the time horizon most commonly used to evaluate mutual fund performances.
Our model does not intend to capture all investment decisions made within the
quarter – stock purchases and sales made early in a quarter may be motivated
by reasons not modelled here. The key is only that the portfolio’s end-of-quarter
composition is influenced by the features that we model, and that better past
performing funds tend to have more cash on hand at the quarter’s end because
of higher accumulated cash inflows throughout the period. We do not claim that
cash inflows received earlier in the period are held in cash (or a passive equity
equivalent); although we do not model it formally, fund managers presumably
initially allocate that cash based on private information and so on and then
reallocate at quarter’s end to reflect the features that we model. This modelling
10 Hu, Hall, and Harvey (2000) find that promotions are positively related to the fund’s past
performance and demotions are negatively related to past performance.
12. 730 D. Bernhardt and R.J. Davies
assumption is consistent with the empirical results of Alexander, Cici, and Gibson
(2007), who show that large fund inflows and outflows influence a fund manager’s
trades. It is also in the same spirit as the theoretical model of Hugonnier and
Kaniel (2008), who examine portfolio choices by a mutual fund manager who
faces dynamic fund flows. They show that proportional fees in combination with
the performance-flow relation can cause a manager to distort a fund’s risk profile.
3.2.4. Definition of fully invested funds
Most mutual funds have small cash holdings that largely reflect (i) recent cash
inflows from investors that have not yet been invested; and (ii) funds held as
an insurance to reduce the transaction costs associated with (stochastic) fund
redemptions. Yan (2006) finds that the average cash holdings of U.S. domestic
equity mutual funds is about 5.3% of assets and shows that the level of cash
holdings is primarily driven by fund flow volatility. In fact, Edelen (1999) finds
that about 70% of all mutual fund trades are to meet investor liquidity demands.
Our model does not incorporate such investor liquidity demands, and hence it
does not contain an insurance role for cash holdings. Consequently, when our
model refers to fully invested funds, we are thinking about real-world funds with
cash positions no larger than those optimally held as insurance against their
expected redemptions.
4. Analysis
4.1. Distortion of investment toward existing holdings
The cost of the fund’s new stock purchases, (I At , I Bt ), at period t are
costAt = [PAt + PAt (PAt , IAt )]IAt (4)
costBt = [PBt + PBt (PBt , IBt )]IBt . (5)
Differences in signals across stocks may offset the incentives to distort investment
toward existing holdings; thus, to control for differences in signals, we reverse the
signals when comparing investments in each stock. Specifically, consider two
opposing signal patterns: (δ A,t+1 = δ 1 , δ B,t+1 = δ 2 ) and (δ A,t+1 = δ 2 , δ B,t+1 =
δ 1 ). Let cost12 denote the cost of optimal investment in stock A under the first
At
signal pattern, and cost21 denote the cost of optimal investment in stock B under
Bt
the second signal pattern, where optimal investments are determined by the fund
manager’s investment problem in (3). Our first result is
PROPOSITION 1. Given assumptions A1–A5, controlling for differences in informa-
tion signals across stocks, the fund manager distorts current purchases toward the
stock in which he has a larger existing position: cost12 > cost21 when PAt SAt >
At Bt
PBt SBt .
13. Smart fund managers? Stupid money? 731
The proof follows directly from the first-order conditions of the fund manager’s
problem. Note that proposition 1 does not rule out the possibility that the fund
manager’s private information signal for stock B is so large that it dominates the
portfolio pumping incentives to purchase stock A. We can provide a stronger
statement about investment in expectation if the distribution of signals is such
that the manager is just as likely to get a relatively strong buy/sell signal for stock
A as we are for stock B. Formally, we say that the joint density distribution of
information signals is symmetric when
g(·, δ B = y) = g(δ A = y, ·), ∀y. (6)
Adding this symmetry condition yields
COROLLARY 1. If the joint density distribution over the fund manager’s private
information signals is symmetric, then in expectation (over possible signals) the
manager invests more money on the stock in which he has a larger existing
position.
Corollary 1 follows immediately from integrating over signals pairwise.
Proposition 1 presents conditions under which the fund manager distorts his
investment toward assets in which he has larger existing positions. The assumed
structure on the price impact of order flow, P(Pjt , Ijt ), ensures that independent
of initial share prices, P At and P Bt , and initial holdings, SAt and SBt , the fund
manager always wants to distort investment toward assets in which he holds larger
positions. The result always holds without this structure (i) if cash inflow is so
high that the fund manager holds some cash or (ii) if, instead, the fund manager is
fully invested in the market, the sufficient condition in proposition 1 always holds
as long as either |P At − P Bt | or |δ 1 − δ 2 | are small enough. Essentially, absent
the structure on P(Pjt , Ijt ), we have to consider second-order effects related to
the relative value of purchases, and these depend on differences in ex ante share
prices and differences in signals.
4.2. The likelihood of information revelation
We now reflect on how the likelihood that the fund manager’s private information
is revealed affects his incentives to distort the fund’s purchases toward existing
holdings. Our next result maintains the same structure on the price impact of
order flow, and adds the condition
PA SA 1 + δA
> . (7)
PB SB 1 + δB
This condition ensures that the difference in existing holdings is large relative to
the difference in information signals.
14. 732 D. Bernhardt and R.J. Davies
LEMMA 1. There exists an > 0 such that condition (7) holds if δ A,t+1 ≤
δ B,t+1 + .
We now derive the implications of condition (7):
PROPOSITION 2. Given assumptions A1–A5, new investment in the fund’s largest
existing holding (stock A) is a decreasing function of the probability γ that infor-
mation about asset-value innovations is revealed before the end of the period if and
only if condition (7) holds.
The proof follows from exploring how investment must change in response to
a change in γ in order to maintain the first-order condition of the manager’s
investment problem.
The intuition is as follows. When condition (7) holds, the manager’s private
information advantage for buying stock A relative to buying stock B is small (or
negative). In this case, the main incentive for the manager to distort investments
toward stock A is a portfolio pumping motive. An increase in the likelihood of
information revelation decreases the value of portfolio pumping, since with a
higher likelihood the end-of-quarter price will reflect fundamental information
rather than the pumping activity. In contrast, when condition (7) does not hold,
the manager has such a large informational advantage for trading stock A that
his primary incentive is to trade aggressively on the basis of this information. In
this case, the more likely this information is to be revealed prior to the end of
the quarter (and thereby incorporated into this period’s returns), the greater is
the value of purchasing stock A, so that investment in stock A is an increasing
function of γ .
Adding symmetry, that is, g(·, δ B = y) = g(δ A = y, ·), we can extend
proposition 2 to consider expected investment, unconditional on the realization
of the signal.
COROLLARY 2. If the joint density distribution over the fund manager’s private
information signals is symmetric and the dispersion of signals is sufficiently small,
then, unconditionally, expected investment in the stock with the larger existing
position declines with the probability that the fund manager’s private information
is revealed.
A smaller γ represents a lower probability of information leakage. This lower
probability can reflect both a smaller period of time between the purchase of
stocks and a period’s end for the information to leak out, and it can reflect the
observation that smaller stocks are less actively followed, so that there are fewer
sources to disseminate the information. Hence, the proposition suggests that fund
managers distort investment more toward existing holdings in smaller stocks or
later in an evaluation period, where price impacts of trades are more likely to
persist.
15. Smart fund managers? Stupid money? 733
4.3. The importance of the cash constraint
We have shown that a fund manager distorts investments toward existing stock
holdings. Despite this distortion, short-run performance may not be persistent.
In particular, we now show that if the fund manager is not fully invested in the
market, then the model cannot generate short-run persistence in fund perfor-
mance.
PROPOSITION 3. Suppose P∗ j,t−1 = Pjt . If the fund manager optimally does not use
all available cash to invest in stocks (and thereby holds cash at the end of the period)
and expected mutual fund returns exceed the risk-free rate, then the mutual fund’s
period returns are a declining function of the cash available to be invested at the
beginning of the period.
The proof follows from recognizing that, when the manager optimally holds a
cash position, the marginal value of relaxing the fund’s budget constraint (the
Lagrange multiplier) is equal to the return on cash.
The expected mutual fund return exceeds the risk-free rate either if the fund
manager has non-negative cash available to invest ( f ≥ 0) and stocks do not
receive large negative signals (δ j > δ j ), or if negative cash inflows are offset by
¯
sufficiently large positive signals. This is because the fund manager can always
earn the risk-free return by investing new fund inflows in cash. In our risk-neutral
setting, it is appropriate to compare mutual fund returns with the risk-free rate.
More generally, the rational behaviour of risk-averse investors implies that ex-
pected mutual fund returns must exceed a comparable risk-adjusted benchmark.
This fact, taken in conjunction with proposition 3, implies that if fund managers
are not fully invested in stocks, then the model is inconsistent with the empirical
regularity that returns exhibit short-run persistence. Indeed, in proposition 3 we
assume that P∗ = Pjt , so that returns are calculated under the assumption that
j,t−1
the end-of-period t − 1 share price reflected the value of the firm at that moment;
that is, there was no past investment distortion. If f (r t−1 ) was higher because
of past investment distortion, this would lead to even more negatively correlated
short-run returns.
4.4. Short-run persistence and long-run reversals
In our model, persistence is best interpreted in terms of the persistence of relative
(ranked) performance: there is short-run persistence when a highly ranked per-
former at t − 1 tends to be highly ranked in t; and lower-ranked performers in t − 1
tend to be lower ranked in t. Because we do not model why a given fund’s return
is initially higher (perhaps better past private information), we cannot compare
the absolute levels of returns from one period to the next, but we can determine
when the factors that we model lead a fund that does relatively better at t − 1 to
do better at t, but eventually to do worse over sufficiently longer horizons.
In the context of our model, long-run returns are just the multi-period return
realized by a sequence of decisions made by a fund manager (or multiple fund
16. 734 D. Bernhardt and R.J. Davies
managers). Because short-run returns eventually fall for f (r t−1 ) sufficiently large,
‘Ponzi-schemes’ cannot be supported in the long run.
Our model does not introduce persistence of fund manager ability. Here, we
focus on the incentives to portfolio pump and show that this portion of the story
generates long-run mean reversion. A reversal in performance means that worst
past performance implies eventual better future performance on a ranking scale.
A fund overperforms by more when its rank is higher over the evaluation period
that we calculate returns; and a fund underperforms by more when its return
rank is lower.
We first show that the model generates short-run persistence in performance
if managers are fully invested in the market. In particular, proposition 4 proves
that, if a fund manager is fully invested, then short-run returns first rise with cash
inflows, f (r t−1 ).
PROPOSITION 4. Suppose P∗ j,t−1 = Pjt . Suppose that the fund manager has the
same private information about each stock (δ A,t+1 = δ B,t+1 = δ) and that he is fully
invested in stocks (i.e., does not hold a cash position). Then there exists an f ∗ such
that, for f < f ∗ , a higher return last period (and hence a higher f) implies a higher
return this period; and for f ≥ f ∗ , further increases in returns in the previous period
imply a lower current return.
In other words, short-run returns first rise with new funds under management
(∂r t /∂ f > 0) for cash inflows sufficiently small, but are a declining function
of new funds for cash inflows sufficiently large. The finding for f < f ∗ is the
relevant one, as it implies performance persistence (past higher return implies
current higher return; i.e., better past performers continue to perform better).
The persistence documented in proposition 4 is driven precisely by the flow-
related buying that pushes up stock prices that Wermers (2003) documents. Ob-
serve that ∂r t /∂ f > 0 when f is small even if a fund manager has no private
information, so that δ = 0. The intuition for proposition 4 is sharpest when
SAt ∼ SBt and P At ∼ P Bt , in which case I At ∼ I Bt ∼ 0. Then Ijt shares are
purchased at a premium of P(Pjt , Ijt ), so the ‘cost’ P(Pjt , Ijt )Ijt is only of
second order. In contrast, the price impact of the share purchase on returns has a
first-order positive impact of order (1 − γ ) P(Pjt , Ijt )Sjt . 11 The rest of the proof
shows that the result extends when a fund manager makes non-trivial offsetting
investments in the two stocks. 12
At date t + 1, δ A,t+1 and δ B,t+1 are revealed and incorporated into prices,
so that, ceteris paribus, greater investment distortions at date t reduce returns
in t + 1 by more. Possibly offsetting this decline is the fact that the period t
11 One can re-interpret proposition 3 in the context of proposition 5 as saying that if it is optimal
for a fund not to be fully invested, then cash inflows are too high relative to private information,
so that short-run returns are a decreasing function of f (r t−1 ).
12 As in proposition 3, we calculate returns using P∗ = Pjt , that is, higher values of f (r t−1 )
j,t−1
primarily reflect factors other than greater past investment distortion.
17. Smart fund managers? Stupid money? 735
investment distortion induced greater cash inflows, f (r t ), which, in turn, can
facilitate another round of investment distortion. To derive the impact of past
performance for long-run returns, one just cumulates short-run returns over time.
While high cash inflows in period t of f (r t−1 ) due to better performance in t − 1
lead to higher returns in period t, for short-run returns in the next period t + 1
not to fall, the second round of investment distortion must dominate the impact
due to the revelation of δ A,t+1 and δ B,t+1 . For longer-run returns to continue to
rise, it must be that the higher short-run return from the immediate distortion of
investment more than offsets lower returns due to realizing past distortions. But
short-run returns must eventually decline with cash inflow, so once cash inflow
is sufficiently high, this cannot occur: eventually, realizing lower returns from
greater past distortions must lead to lower long-run returns.
We now consider a fund manager who has no private information, so that
greater incremental investments in existing positions represent more costly dis-
tortions that must eventually be realized in the form of lower returns. It follows
that greater cash inflows in period t must lead to lower long-run returns when
cumulated over the period [t, t + τ ], for τ sufficiently large. Combining this
observation with proposition 4, yields the following proposition, documenting
that the model reconciles both the short-run persistence in fund performance
and the long-run reversal in fund performance documented by Zheng (1999) and
Wermers (2003).
PROPOSITION 5. Even if a fund manager has no private information, for f (r t−1 )
sufficiently small, short-run returns are a rising function of f (r t−1 ), but long-run
returns over [t, t + τ ] are a declining function of f (r t−1 ), for τ sufficiently large.
Proposition 5 indicates that a better performer in t − 1 will be higher ranked in
t (on the basis of returns), but lower ranked (on the basis of returns) by some
date τ .
Because the fund manager has no private information, from a long-run per-
spective, the optimal stock investment is zero. For τ sufficiently high, long-run
returns are a strictly declining function of f (r t−1 ), because the price premium
paid for each share rises with the investment. We show later that τ need not be
very large for long-run returns to fall with f (r t−1 ).
These results hold even if past returns were so bad that redemptions lead to
a net outflow of money from the fund. Then the mutual fund manager must
disinvest. To minimize the adverse return consequences the fund manager tends
to sell stocks in which he has smaller positions (i.e., the negative price impact from
selling impacts a smaller proportion of his portfolio). Further, in the environment
characterized by proposition 5, short-run returns are lower (and negative) for
fund managers with greater redemptions, but there will be a long-run reversal in
performance. The intuition is analogous to that described before: in this situation,
a fund manager sells near the end of the quarter, but then begins the next quarter
with a positive return (as prices revert back to fundamental value); eventually,
18. 736 D. Bernhardt and R.J. Davies
as more quarters pass, this positive return is large enough to offset the negative
impact of fund outflows, thereby allowing the fund to become a ‘winner’. In
general, the loser-to-winner effect will be smaller than the reverse effect, since the
stocks being sold constitute a smaller proportion of the fund’s holdings.
Because cash inflows are more sensitive to fund performance for newer funds
(Chevalier and Ellison 1997), the model predicts that the persistence in short-run
returns should decline as funds mature, and, in turn, there should be a smaller
long-run reversal in performance for mature funds, as Zheng (1999) documents.
Over time, funds with high short-run returns should underperform the market in
the long run, as Zheng also finds.
Are mutual fund investors stupid? Our theoretical analysis and Zheng’s em-
pirical work suggest that it is advisable to invest in a fund that just performed
well for the ‘first’ time. Our results indicate that if investors reassess their mutual
fund holding on a quarterly basis, then the optimal time to exit a fund is related
to both the previous quarter’s performance and the length of time the fund has
been performing well; the more time passes, the higher the previous quarter’s
performance must be to merit continuing to hold the fund.
5. Examples
Using a series of simple examples, we now illustrate how the return-cash flow
relationship characterized above emerges in the short- and long-run.
5.1. Identical stock holdings
Consider an economy in which the fund manager has no private information,
δ A,t+1 = δ B,t+1 = 0, and identical holdings, SAt = SBt = S > 0. The two identical
stocks share an initial common price of P At = P Bt = 1, and trades have a linear
price impact, P(1, Ijt ) = a I jt , a > 0.
It is straightforward to verify that there is a critical value f¯ such that the fund
manager is fully invested if and only if cash inflows, f , are less than f¯. Further,
for f < f¯, the fund manager optimally divides his purchases equally between
√
stocks, purchasing I( f ) = 1 + 2a f − 1/2a shares of each stock to generate an
expected mutual fund return of
2(I( f ) + S)(1 + (1 − γ )a I( f )) − (2S + f )
E[r ( f )] = .
2S + f
Initially, returns from investing are increasing in f ,
∂ a(1 − γ )
[E[r ( f = 0)]] = > 0,
∂f 2
but the second derivative with respect to f is negative,
19. Smart fund managers? Stupid money? 737
FIGURE 2 Fund flow-return relationship in base example. Parameters: SAt = SBt = 20, a = b =
0.02, γ = 0.5, P At = P Bt = 1, δ A,t+1 = δ B,t+1 = δ
∂2 a[−(1 − γ )S − 1]
[E[r ( f = 0)]] = < 0.
∂f2 2S
Figure 2 illustrates the fund flow-return relationship. The short-run portfolio
return first rises and then falls with cash flow into the fund. But note that the
cash inflow that maximizes short-run returns, f¯, is quite high, about 40% of
the portfolio value. That is, although proposition 4 proves only that short-run
returns are initially increasing in cash inflow, the figure reveals that short-run
returns can continue to rise even if f is substantial. Hence, there is short-run
persistence in portfolio performance, as better immediate past performance draws
more cash inflow, which gives rise to higher current returns. To the extent that
fund managers differ in ability or there are momentum effects 13 the short-run
persistence is further magnified. 14
Within this symmetric framework, a comparative statics analysis yields sharp
predictions:
13 Carhart (1997), Sapp and Tiwari (2003), and Wermers (2003) find that winning funds hold
winning stocks, while the worst-performing funds tend to hold the largest positions in the
worst-performing stocks. Our model predicts that funds are particularly reluctant to sell their
largest positions (as Wermers 2003 finds), so that any momentum effects would magnify the
persistence in short-run performance and delay the long-term performance reversal in our
model. Still, Wermers finds that such momentum effects explain only a small portion of the
persistence (price impact from fund purchases explain more).
14 We can extend our model to consider large fund complexes in which a fund manager of a large
fund within the fund family may have the incentive to buy shares of stocks in which a smaller
fund within the fund family has a large stake. The cost of the newly purchased shares has a
negligible effect on the returns of the large fund, but could have a large positive effect on the
small fund’s return. Given the documented return-flow relationship, the net result would be
higher net cash inflows for the fund family. If this practice is widespread, we would expect to
observe short-term persistence and long-term reversals in the aggregate returns of a family of
funds.
20. 738 D. Bernhardt and R.J. Davies
• Short-run returns fall with the information arrival rate, γ . That is, short-run
returns are higher if information is less likely to leak out by the end of the
period:
∂ E[r ( f )] −2a I( f )(I( f ) + S)
= < 0.
∂γ 2S + f
• Greater cash inflows raise short-run returns by more if information is less likely
to leak out:
∂ 2 E[r ( f )] ∂ I( f ) ∂ I( f )
= − 4a I + 2aS (2S + f )−1
∂γ ∂ f ∂f ∂f
+ 2a I( f )(I( f ) + S)(2S + f )−2
∂ I( f )
= −2a(2S + f )−2 I 2(2S + f ) − I( f )
∂f
∂ I( f )
+ S (2S + f ) − I( f ) < 0,
∂f
because
∂ I( f )
f > I( f ) and S ≥ 0.
∂f
These two predictions can reconcile the finding of Zheng (1999) and others
that niche mutual funds (which invest in stocks where there are fewer sources of
information) have stronger short-run persistence and greater long-run reversals
in performance.
5.2. Differences in price schedules
We now explore how outcomes are affected if the price impact of trades differs
across stocks. We suppose that P At (I At ) = a I At and P Bt (I Bt ) = bI Bt , where
a = b, but that the stocks are otherwise identical, P At = P Bt = 1 and δ At =
δ Bt = δ. Figure 3 numerically characterizes how I At depends on a when the fund
manager is fully invested. We consider two cases: (i) balanced holdings (SAt =
SBt = S); and (ii) unbalanced holdings (SAt > SBt ). The figure highlights that
as long as cash inflow, f , is sufficiently small, investment in stock A rises with
a; but for higher values of f , investment falls with a. Further, these investment
consequences are magnified when SAt > SBt . Phrased differently, if and only if
cash inflows are not too high, the gain from strategically manipulating returns
by distorting investments toward greater existing positions rises with the price
impact of stock A order flow, a. However, if cash inflows are too high, increases
in a reduce investment in stock A, since the strategic manipulation gains are
dominated by the increasing marginal cost of additional share purchases.
21. Smart fund managers? Stupid money? 739
FIGURE 3 Fund manager’s investment decision and differences in price sensitivity. Parameters:
γ = 0.5, P At = P Bt = 1, δ A,t+1 = δ B,t+1 = 0.07, b = 0.02. Balanced corresponds to SAt = SBt = 20.
Unbalanced corresponds to SAt = 30 and SBt = 10. For both the balanced and unbalanced cases, we
plot the fund manager’s investment in stock A, I At , versus the price impact of stock A to order flow
(a) for low cash inflows ( f = 4) and for high cash inflows ( f = 18)
Next, we use our framework to predict differences in the investment behaviour
of specialized funds. To do so, we equate the price impact of order flow for the two
stocks (a = b), and then consider how investment, stock prices, and fund returns
change as we vary this parameter. Figure 4 illustrates outcomes when SAt = 30
and SBt = 10. Since the only distinction between the two stocks is the existing
holdings, without the short-term painting the tape motive, the long-term optimal
investment would be to purchase the same amount of both stocks. Panel A shows
that as stock price sensitivity increases, the amount of investment distortion (as
measured by the difference between I At and I Bt ) falls. At the same time, panel B
shows that the price movement in the more sensitive (less liquid) stocks is larger.
Thus, our model predicts that a fund with holdings in small, illiquid stocks will
have greater return distortions, but less investment distortion, than an otherwise
comparable fund with holdings in larger, more liquid stocks. As well, the apparent
clairvoyant stock-picking effect should be reduced for larger, more liquid stocks,
since these stocks will have less price impact from distortionary trades.
5.3. Returns over multiple periods, different initial stock holdings
The numerical findings illustrated in figure 2 do not imply that higher returns
will persist for some time. This is because a fund must realize the negative re-
turn consequences of its immediate past investment distortions in each period,
as share prices incorporate the true value of past signals. Figure 5 considers
22. 740 D. Bernhardt and R.J. Davies
FIGURE 4 Investment behaviour of specialized funds that focus on stocks with a given price
sensitivity. Parameters: γ = 0.5, P At = P Bt = 1, δ A,t+1 = δ B,t+1 = 0.07, f = 4, SAt = 30 and SBt = 10.
Both stocks have the same price sensitivity (a = b). The fund is fully invested in stocks for the
parameters considered. The expected price change for stock j is [γ a (Pjt Ijt ) + (1 − γ )δ j,t+1 ] Pjt .
longer-period returns when cash inflows in each period t are given by f (r t−1 ) =
[0.2 + 4.2(r t−1 − δ)] × (P A,t−1 SA,t−1 + P B,t−1 SB,t−1 + Mt−1 ), where δ = δ At =
δ Bt ∀t. This flow-performance relation is a linear approximation of the relation
estimated by Chevalier and Ellison (1997) for two-year old mutual funds, and
subsequently modelled by Berk and Green (2004). We illustrate period 1 fund
flows ranging from −10% to 40% of existing holdings, which is roughly con-
sistent with the typical range of annual fund flows observed by Chevalier and
Ellison (1997) for two-year old mutual funds.
The figure highlights that better-performing funds in period 0, which conse-
quently generate high cash inflows at the beginning of period 1, also do better
in period 1: there is very strong short-run persistence in performance. However,
in period 2, past better performers underperform by amounts that substantially
offset their superior period 1 performance, and were we to continue for another
23. Smart fund managers? Stupid money? 741
FIGURE 5 Multi-period example illustrating short-term persistence and long-run reversal of
performance. Period t cash inflow is calculated as: f t (r t−1 ) = [0.2 + 4.2 (r t−1 − δ)] × (P A,t−1 SA,t−1 +
P B,t−1 SB,t−1 + Mt−1 ). Period 0 winners are funds for which r 0 > i 0 . The y-axis plots mutual fund
returns (r t , t = 1, 2) and the x-axis plots period 1 cash inflows f 1 (r 0 ). Parameters: a = b = 0.02,
SA,1 = 30, SB,1 = 10, P A,1 = P B,1 = 1, γ = 0.5; δ A,t = δ B,t = δ = 0.07 ∀t
period, there would be a long-run reversal in performance. Obviously, if we in-
corporated other persistent factors, such as fund manager ability, then long-run
reversals would take longer and be less extreme.
According to our flow-performance relation, the initial range of cash inflows
(−10% to 40%) implies a difference in period 0 returns of 0.3% to 30.0%. This
dispersion of returns is much larger than the dispersion of observed period 1 re-
turns obtained using our model: 5.7% to 12.5%. This suggests that, while painting
the tape drives short-run persistence and long-run reversal patterns, other exoge-
nous factors such as idiosyncratic stochastic shocks drive much of the dispersion
in observed returns. In particular, this confirms that taking r 0 and hence initial
cash flows as exogenous (as we do in our propositions) is unimportant, as our
qualitative findings would extend. 15
There is significant anecdotal and empirical evidence that suggests funds con-
duct much of their trade near the end of a quarter. In addition to the trad-
ing behaviour that we model, there is evidence of ‘window dressing’ (selling the
fund’s lacklustre holdings before they are revealed to investors in end-of-quarter
reports). If selling poorly performing stock holdings near the end of quarters is
prevalent, then fund managers will have more cash on hand than normal to make
new investments near the end of quarters. Our model predicts that fund managers
will tend to use this cash to buy shares that magnify the value of the fund’s ex-
isting holdings (i.e., stocks in which the fund has many shares or stocks that are
15 In unreported results, we confirmed that our qualitative findings are not affected by starting
values by iterating our model over an arbitrary number of periods and verifying that the same
short-run persistence and long-run reversal patterns exist in later periods.
24. 742 D. Bernhardt and R.J. Davies
thinly traded). This behaviour has very different implications for mutual fund
return patterns than the window-dressing motive of buying winners (in which
a fund may or may not have a substantial existing position). Window dressing
alone cannot generate the short-run persistence and long-run reversal patterns
in the data. 16
6. Conclusion
This paper provides a theoretical framework for understanding the incentives for,
and the implications of, mutual fund managers distorting end-of-quarter stock
purchases toward stocks in which the fund has larger existing positions. We show
how this common practice, sometimes known as painting the tape, depends on the
short-term price impact of trades and the relation between past fund performance
and future fund cash inflows.
We derive when and how such trading leads to the empirically observed short-
run persistence and long-run reversal in fund performance. Our model also ex-
plains why mutual funds tend to be relatively undiversified and exhibit persistent
stock selection. Our theoretical predictions are consistent with the empirical find-
ings of Carhart et al. (2002) that trading-induced equity price inflation on the last
day of a quarter gives rise to abnormal fund returns on those days and that the end
of year performance effect is more pronounced for better historical performers.
Our unified framework for analyzing mutual fund investment distortion pro-
vides several directions for future empirical research. For instance, fund managers
will invest more heavily on the basis of private information earlier in an evaluation
period and this should be reflected in the long-run return characteristics of these
purchases. As well, the short-run persistence, long-run reversal pattern should be
exhibited only by funds that are effectively fully invested in the market (excluding
cash held for insurance against unexpected redemptions). And funds with a more
sensitive performance-flow relation should exhibit greater investment distortion.
Finally, we argue that existing measures of mutual fund managers’ true stock
picking ability have been understated because of their failure to recognize the
additional costs of the strategic behaviour modelled in our paper.
Appendix A: Proofs
We note that maximizing portfolio returns also maximizes the portfolio end-of-
period value. Thus, the Lagrangian corresponding to the manager’s problem is
16 So, too, Carhart et al. (2002) find that mutual fund net asset value inflation in excess of the S&P
500 index on the last day of a quarter is at least 0.2% to 0.4% for funds in lesser deciles. Carhart
et al. argue that this is caused by a ‘leaning-for-the-tape’ motive, in which fund managers near
the top of the return rankings have the most to gain from performance rank improvements
(consistent with the empirical findings of Ippolito 1992 and Sirri and Tufano 1998) and thus
have the most to gain from end-of-quarter distortionary trades. But this leaning-for-the-tape
motive alone cannot generate short-term persistence in mutual fund returns because there is no
link between past- and current-period fund performance.
25. Smart fund managers? Stupid money? 743
L(Mt , IAt , IBt ) = Mt + (I j t + Sj t )E Pj∗t + λ M Mt + λ A SAt + IAt
j
+ λ B SBt + IBt + λbud f (rt−1 ) + Mt−1 − [PAt
+ PAt (PAt , IAt )]IAt − [PBt + PBt (PBt , IBt )]IBt − Mt .
Given that λ M = λ A = λ B = 0, the associated first-order condition characterizing
the optimal investment in stock j ∈ {A, B} is
∂L ∂ Pj t
= Pj t + (1 − γ ) Pj t (Pj t , I j t ) + γ δ Pj t + (1 − γ )I j t
∂ Ijt ∂ Ijt
∂ Pj t ∂ Pj t
+ (1 − γ )Sj t − λbud Pj t + P(Pj t , I j t ) + I j t = 0.
∂ Ijt ∂ Ijt
Proof of proposition 1. From the first-order conditions:
∂ P
PAt + (1 − γ ) P(PAt , IAt ) + γ δ1 PAt + (1 − γ )(IAt + SAt )
∂ IAt
λbud = (8)
∂ P
PAt + P(PAt , IAt ) + IAt
∂ IAt
∂ P
PBt + (1 − γ ) P(PBt , IBt ) + γ δ2 PBt + (1 − γ )(IBt + SBt )
∂ IBt
= . (9)
∂ P
PBt + P(PBt , IBt ) + IBt
∂ IBt
Substituting for P(P j , I j ) = k (Pj Ij ) P j ,
γ + γ δ1 + PAt SAt (1 − γ )k γ + γ δ2 + PBt SBt (1 − γ )k
(1 − γ ) + = (1 − γ ) + .
1 + 2kPAt IAt 1 + 2kPBt IBt
(10)
Then, reversing PAt SAt and PBt SBt , holding PAt IAt and PBt IBt fixed, it is clear that
γ + γ δ1 + PBt SBt (1 − γ )k γ + γ δ2 + PAt SAt (1 − γ )k
< .
1 + 2kPAt IAt 1 + 2kPBt IBt
To restore equality, it follows that the optimal values of I At and I Bt must satisfy
PAt IAt (δ 1 , δ 2 ) > PBt IBt (δ 2 , δ 1 ). Then it follows that
(PAt + P(PAt , IAt (δ1 , δ2 )))IAt (δ1 , δ2 )
> (PBt + P(PBt , IBt (δ2 , δ1 )))IBt (δ2 , δ1 ).
Proof of proposition 2. Re-arranging (10) gives
γ (1 + δ A − kPA SA) + kPA SA 1 + 2kPAt IAt
= . (11)
γ (1 + δ B − kPB SB ) + kPB SB 1 + 2kPBt IBt
26. 744 D. Bernhardt and R.J. Davies
Taking the derivative of the left-hand side of (11) with respect to γ gives
k(PB SB − PA SA) + δ A PB SB k − δ B PA SAk
.
[γ (1 + δ B − kPB SB ) + kPB SB ]2
Since PA SA > PB SB , this derivative is negative if δ A ≤ δ B or more generally iff
condition (7) holds. Since the right-hand side of (11) is increasing in PAt IAt , our
desired result follows directly.
Proof of proposition 3. If Mt > 0, then λ bud = 1 and the marginal dollar is invested
in cash (dM t /d f = 1; ∂ Ijt /d f = 0, j = A, B). Differentiating short-run expected
returns,
∗ ∗
E[rt ] = Mt + E[PAt ](IAt + SAt ) + E[PBt ](IBt + SBt )
−1
× PAt SAt + PBt SBt + f (rt−1 ) − 1,
with respect to f = f (r t−1 ) at the manager’s optimal values of I At and I Bt yields
∂ E[rt (IAt , IBt )] d Mt d ∗ ∗
= + E[PAt ](IAt + SAt ) + E[PBt ](IBt + SBt )
∂f df df
∗
× [PAt SAt + PBt SBt + f ]−1 − [Mt + E[PAt ](IAt + SAt )
∗
+E[PBt ](IBt + SBt )][PAt SAt + PBt SBt + f ]−2
∗
= [PAt SAt + PBt SBt + f ]−1 − [Mt + E[PAt ](IAt + SAt )
∗
+ E[PBt ](IBt + SBt )][PAt SAt + PBt SBt + f ]−2
= −E[rt ][PAt SAt + PBt SBt + f ]−1 ≤ 0,
if E[rt ] ≥ 0; strict if E[rt ] > 0,
where the last equality follows from substitution.
Proof of proposition 4. From the budget constraint,
f = [PAt + PAt (PAt , IAt )]IAt + [PBt + PBt (PBt , IBt )]IBt (12)
∂ IAt ∂ PAt ∂ IAt ∂ IBt
1= (PAt + PAt ) + IAt + (PBt + PBt )
∂f ∂ IAt ∂ f ∂f
∂ PBt ∂ IBt
+ IBt .
∂ IBt ∂ f
We first prove the result that short-run returns are increasing in f when I At and
I Bt are small if f is close to zero (as will be the case for SAt ∼ SBt , and P At ∼ P Bt ).
27. Smart fund managers? Stupid money? 745
Then, it follows that
∂ IAt ∂ IBt
1= PAt + PBt .
∂f ∂f
If the fund manager is fully invested, then expected returns are
E[rt ] =
(1 + γ δ)Pj t (I j t + Sj t ) + (1 − γ ) Pj t (Pj t , I j t )(I j t + Sj t )
j ={A,B}
− 1.
PAt SAt + PBt SBt + f (13)
Thus,
∂ E[rt ] ∂ PAt ∂ IAt
= [PAt SAt + PBt SBt ]−1 (1 − γ ) SAt
∂f f =0 ∂ IAt ∂ f
∂ PBt ∂ IBt ∂ IAt ∂ IBt
+ SBt + (1 + γ δ) PAt + PBt
∂ IBt ∂ f ∂f ∂f
(1 + γ δ)(PAt SAt + PBt SBt )
− .
(PAt SAt + PBt SBt )2 (14)
Substituting for P Bt (∂ I Bt /∂ f ) + P At (∂ I At /∂ f ) = 1 yields
∂ E[rt ] ∂ PAt ∂ IAt
= [PAt SAt + PBt SBt ]−1 (1 − γ ) SAt
∂f f =0 ∂ IAt ∂ f
∂ PBt ∂ IBt
+ SBt > 0, (15)
∂ IBt ∂ f
since ∂ P kt /∂ I k > 0 (by assumption) and ∂ I kt /∂ f > 0 (since (10) can hold only
if an increase in I At coincides with an increase in I Bt ).
More generally to show that the derivative of returns with respect to f is
positive recognize that this amounts to showing that
∂num
sign (PAt SAt + PBt SBt ) − num > 0,
∂f | f =0
where num is the numerator of expected returns. Equivalently, dividing through
by PAt SAt + PBt SBt , it amounts to showing that
∂num
sign − (1 + E[rt ]) = sign λbud − (1 + E[rt ]) > 0,
∂f | f =0 | f =0
where the equality follows from substitution of equations (8) and (9). That is,
λ bud reflects the marginal return of one more dollar, and if the marginal return
28. 746 D. Bernhardt and R.J. Davies
exceeds 1 + r t , then r t must be increasing in f . The Lagrange multiplier can be
interpreted as the marginal value of one more dollar of investment in stock A or
as the reduction in the marginal cost of selling one more dollar of investment in
stock B. We now show that for δ ≥ 0, this marginal cost grows monotonically as
the fund manager sells more I B . Differentiating (9) with respect to I Bt , we see
that it has sign
(PBt + P(PBt , IBt ) + IBt P (PBt , IBt ))
× [(1 − γ )(2 P (PBt , IBt ) + IBt P (PBt , IBt )) + SBt (1 − γ ) P (PBt , IBt )]
−(2 P (PBt , IBt ) + IBt P (PBt , IBt ))
× [PBt + (1 − γ ) P(PBt , IBt ) + γ δ PBt + (1 − γ )(IBt + SBt ) P (PBt , IBt )]
< (PBt + P(PBt , IBt ) + IBt P (PBt , IBt ))SBt (1 − γ ) P (PBt , IBt )]
− (2 P (PBt , IBt ) + IBt P (PBt , IBt ))(γ δ PBt + SBt (1 − γ ) P (PBt , IBt ))]
< 0,
where the last inequality follows from δ ≥ 0 and A4. λ bud takes its value evaluated
at I ∗ , so that it exceeds the derivative evaluated at I Bt = 0, which we have proven
Bt
exceeds r t . For δ < 0 an analogous exercise on (8) establishes monotonicity for
I At > 0.
Finally, short-run returns must eventually decline with f : A4 implies that
marginal returns from making arbitrarily large purchases of a stock must
eventually become negative. This implies that once f grows sufficiently large,
the fund manager begins to invest in cash at the point where the marginal
return on investing in stock is i (which we normalized to zero), and where the
short-run portfolio return exceeds i. It follows that for f larger, short-run returns
decline.
References
Alexander, G.J., G. Cici, and S. Gibson (2007) ‘Does motivation matter when assessing
trade performance? An analysis of mutual funds,’ Review of Financial Studies 20, 125–
50
Baks, K.P. (2007) ‘On the performance of mutual fund managers,’ Journal of Finance,
forthcoming
Berk, J.B., and R.C. Green (2004) ‘Mutual fund flows and performance in rational mar-
kets,’ Journal of Political Economy 112, 1269–95
Bernhardt, D., and R.J. Davies (2005) ‘Painting the tape: aggregate evidence,’ Economics
Letters 89, 306–11
Bernhardt, D., and E. Nosal (2008) ‘Gambling for dollars: a strategic analysis of hedge
funds,’ working paper, University of Illinois
Bernhardt, D., and B. Taub (2008) ‘Cross-asset speculation in stock markets,’ Journal of
Finance, forthcoming
Bhattacharyya, S., and V. Nanda (2007) ‘Portfolio pumping, trading activity, and fund
performance,’ working paper, University of Michigan
29. Smart fund managers? Stupid money? 747
Bollen, N.P.B., and J.A. Busse (2005) ‘Short-term persistence in mutual fund performance,’
Review of Financial Studies 18, 569–97
Brown, S.J., and W.N. Goetzmann (1995) ‘Performance persistence,’ Journal of Finance
50, 679–98
Caball´ , J., and M. Krishnan (1994) ‘Imperfect competition in a multi-security market
e
with risk neutrality,’ Econometrica 62, 695–704
Carhart, M.M. (1997) ‘On persistence in mutual fund performance,’ Journal of Finance
52, 57–92
Carhart, M.M., R. Kaniel, D.K. Musto, and A.V. Reed (2002) ‘Leaning for the tape:
evidence of gaming behavior in equity mutual funds,’ Journal of Finance 57, 661–93
Chan, L.K.C., and J. Lakonishok (1995) ‘The behavior of stock prices around institutional
trades,’ Journal of Finance 50, 1147–74
Chevalier, J., and G. Ellison (1997) ‘Risk taking by mutual funds as a response to incen-
tives,’ Journal of Political Economy 105, 1167–200
Edelen, R. (1999) ‘Investor flows and the assessed performance of open-end mutual funds,’
Journal Financial Economics 53, 439–66
Gallagher, D.R., P. Gardner, and P.L. Swan (2007) ‘Portfolio pumping: an examination of
investment manager quarter-end trading and impact on performance,’ working paper,
University of New South Wales
Gruber, M.J. (1996) ‘Another puzzle: the growth in actively managed mutual funds,’
Journal of Finance 51, 783–810
Hendershott, T., and M. Seasholes (2007) ‘Market maker inventories and stock prices,’
American Economic Review (Papers and Proceedings) 97, 210–14
Hendricks, D., J. Patel, and R. Zeckhauser (1993) ‘Hot hands in mutual funds: short-run
persistence of relative performance, 1974–1988,’ Journal of Finance 48, 93–130
Hu, F., A. Hall, and C.R. Harvey (2000) ‘Promotion or Demotion? An empirical investi-
gation of the determinants of top mutual fund manager change,’ working paper, Duke
University
Hugonnier, J., and R. Kaniel (2008) ‘Mutual fund portfolio choice in the presence of
dynamic flows,’ Mathematical Finance, forthcoming
Ippolito, R. (1992) ‘Consumer reaction to measures of poor quality: evidence from the
mutual fund industry,’ Journal of Law and Economics 35, 45–70
Kacperczyk, M., C. Sialm, and L. Zheng (2005) ‘On the industry concentration of actively
managed equity mutual funds,’ Journal of Finance 60, 1983–2011
Keim, D.B., and A.N. Madhavan (1996) ‘The upstairs market for large-block transactions:
analysis and measurement of price effects,’ Review of Financial Studies 9, 1–36
— (1997) ‘Transaction costs and investment style: an inter-exchange analysis of institu-
tional equity trades,’ Journal of Financial Economics 46, 265–92
Kyle, A. (1985) ‘Continuous auctions and insider trading,’ Econometrica 53, 1315–35
Malkiel, B.G. (1995) ‘Returns from investing in equity mutual funds, 1971 to 1991,’ Journal
of Finance 50, 549–72
Pinnuck, M. (2003) ‘An examination of the performance of the trades and stockholdings
of fund managers: further evidence,’ Journal of Financial and Quantitative Analysis 38,
811–28
San, G. (2007) ‘Who gains more by trading – institutions or individuals?’ Working paper,
Hebrew University of Jerusalem
Sapp, T.R.A., and A. Tiwari (2004) ‘Does stock return momentum explain the ‘smart
money’ Effect?’ Journal of Finance 59, 2605–22
Sirri, E.R., and P. Tufano (1998) ‘Costly search and mutual fund flows,’ Journal of Finance
53, 1589–622
Wermers, R. (2003) ‘Is money really ‘‘smart’’? New evidence on the relation between
mutual fund flows, manager behavior, and performance persistence,’ working paper,
University of Maryland
30. 748 D. Bernhardt and R.J. Davies
Wermers, R., L. Chen, and N. Jegadeesh (2000) ‘The value of active mutual fund man-
agement: an examination of the stockholdings and trades of fund managers,’ Journal
of Financial and Quantitative Analysis 35, 343–68
Yan, X. (2006) ‘The determinants and implications of mutual fund cash holdings: theory
and evidence,’ Financial Management 35, 69–91
Zheng, L. (1999) ‘Is money smart? A study of mutual fund investors’ fund selection ability,’
Journal of Finance 54, 901–33